Cerebras and OpenAI preview Ultrafast Mode for GPT-5.6 Sol

Cerebras and OpenAI have shared an early look at Ultrafast Mode, a new service tier that launches first in the OpenAI API and is powered by Cerebras. Access opens initially to a select group of customers and expands over time. In this setup Cerebras serves GPT-5.6 Sol, delivering up to 750 output tokens per second — which the companies say comes without any quality compromise.

Ending the speed-versus-intelligence tradeoff

Builders have long had to choose between speed and intelligence: as models scale, they cost more computation and data movement, slowing responses. According to the companies, Sol in Ultrafast mode removes that tradeoff. Compared with output speeds reported by Artificial Analysis, it runs 11 times faster than Fable 5 and 5 times faster than Opus 4.8 in Fast mode.

Humanity's Last Exam in one working day

Cerebras benchmarked the tier on Humanity's Last Exam, a set of 2,500 questions that typically only PhD-level experts can answer. GPT-5.6 Sol in Ultrafast mode worked through all 2,500 questions in 11 hours and 11 minutes; Claude Fable 5 needed 78 hours and 27 minutes — more than three days of continuous compute — for comparable accuracy, nearly 7× slower. Cerebras ran Sol Ultrafast with Codex on xhigh reasoning on July 10, and Fable 5 with Claude Code on July 13–15.

Economically valuable work

On GDP-Val, a benchmark for economically valuable knowledge work, Ultrafast delivered a 5.6x end-to-end speedup with no quality degradation (measured July 31 using medium reasoning inside Codex). OpenAI positions GPT-5.6 Sol as its strongest model yet for legal briefs, financial models and engineering reports — work that is time-sensitive and, as agents move onto the critical path, increasingly speed-bound.